need a compaction strategy for hot/cold data

1、频繁的lowlevelscan触发compact，降低levelt层数，减少key比较，提升读性能；
2、冷数据master触发compact，删除冗余数据，压缩空间；

No files or tests are named. Start by tracing the compaction paths associated with frequent lowlevelscan activity and master-triggered compaction, then determine how hot and cold data should be identified. Done means an agreed compaction strategy that reduces level depth and key comparisons for hot data while removing redundant cold data and reducing space usage.

Written by the indexing model from the issue text.